834127.fig.006a
(a)
834127.fig.006b
(b)
834127.fig.006c
(c)
Figure 6: Abs. training set size versus test set accuracy for gradient features on MNIST, USPS, and DIGITS (SVM variants).